Cross-Border Banking and the International Transmission of Financial Distress During the Crisis of 2007-2008

Cross-Border Banking and the International Transmission of Financial Distress during the Crisis of 2007-2008

Alexander Popovy Gregory F. Udell European Central Bank Indiana University

January 2010

Abstract

We study the e¤ect of …nancial distress in foreign parent banks on local SME …nancing in 14 central and eastern European countries during the early stages of the 2007-2008 …nancial crisis. We use survey data on applicant and non-applicant …rms that enable us to disentangle e¤ects driven by shocks to the banking system from recession-driven demand shocks that may vary across lenders. We …nd strong evidence that credit tightened in the relatively early stages of the crises caused by the following types of bank …nancial distress: 1) low equity ratio; 2) low Tier 1 capital ratio; and 3) losses on …nancial assets. We also …nd that foreign banks transmit to Main Street a larger portion of similar …nancial shocks than domestic banks. The observed decline in credit is greater among informationally opaque …rms and …rms with fewer tangible assets. JEL classi…cation: E44, E51, F34, G21 Keywords: credit crunch, …nancial crisis, bank lending channel, business lending

We thank Santiago Carbo-Valverde, Nicola Cetorelli, Nandini Gupta, Florian Heider, Sebnem Kalemli-Ozcan, Steven Ongena, and seminar participants at the Bank of Finland, the European Central Bank, and the Stockholm School of Economics for useful discussions, as well as Dana Scha¤er and Francesca Fabbri for outstanding research assistance. The opinions expressed herein are those of the authors and do not necessarily re‡ect those of the ECB or the Eurosystem. yCorresponding author. European Central Bank, Financial Research Division, Kaiserstrasse 29, D-60311 Frank- furt, email: [email protected] 1 Introduction

The increasing integration of the European banking industry o¤ers the prospect of important gains in terms of e¢ ciency and diversi…cation, but it also creates potential risks. One such risk is associated with the possibility that a shock to a cross-border bank’s capital will result in a reduction in lending to …rms and consumers in an economic environment that is uncorrelated with the origins of that shock. Given the size and penetration of a number of west European and U.S. banks in central and eastern Europe, their …nancial distress associated with the meltdown of sub- prime mortgages and securitized products in 2007 and 2008 and the run on banks by short-term creditors, counterparties, and borrowers concerned about the liquidity and solvency of the banking sector1, may have led to such a realization.2 The goal of this paper is to put this hypothesis to the test.

We investigate one key mechanism through which foreign …nancial distress may have been transmitted to local economic conditions, namely the supply of credit to small and medium enterprises.

SMEs dominate the corporate landscape in central and eastern Europe, comprising up to 99% of all …rms. Moreover, because of their opacity SMEs may be particularly vulnerable to contractions in the supply of credit. With this high dependency on the SME sector and with immature capital markets, banks are by far the main provider of funds for capital investment and expansion. An important feature of the central and eastern European banking market is its ownership structure.

In particular, foreign ownership in the banking sector has grown so dramatically in the recent decade, that by 2008 foreign banks controlled around 80% of the assets in the the region’sbanking industry. The serious …nancial distress of pan-European banks like Erste, KBC, and Societe Gen- erale since 2007 stemming from economic circumstances unrelated to their operation in central and eastern Europe provides a natural experiment to study the channels through which the e¤ects of

1 See Brunnermeier (2009), Gorton (2009), and Ivashina and Scharfstein (2009) for a timeline of the 2007-2008 global …nancial crisis. See Table 1 for developments concerning the …nancial sector in the countries covered by this paper. 2 Signs of the negative e¤ects of the global …nancial crisis on business …rms in emerging Europe through the channel of bank lending were seen as early as the Fall of 2007. For instance, in October, the EBRD’s chief economist Erik Berglof warned that "the crisis in the West will be a serious one which will last for some time and this means it will de…nitely have an impact on our countries [...] due to the di¢ culties and higher costs associated with obtaining credit" (EBRD (2007)). The euro zone Bank Lending Survey indicated that euro zone banks started tightening lending standards in Q3:2007 (ECB (2008)).

1 the …nancial crisis that started in the U.S. spread through out the global economy.

Our key data come from a survey of a large group of SMEs in emerging Europe administered in April 2005 and April 2008. The data allow us to directly observe …rms whose loan application was turned down over the course of the previous year, or which were discouraged from applying for bank credit by high rates and unfavorable collateral requirements. While we do not observe the bank which granted/denied the loan, we observe the extent of the operations of all banks present in the …rm’s city of incorporation. By using balance sheet data on the parent banks, foreign or domestic, we construct an index of …nancial distress at the level of each locality in 14 countries in the region, which we then map into data on loan rejection rates. The …nal data consist of 4; 421

…rms in 1; 266 localities served by a total of 141 banks over the 2005-2008 period. The majority of localities, however, are served by just a handful of banks, with foreign ownership of those varying by country and locality. This allows us to answer two important questions: 1) did banks transmit their

…nancial distress by shrinking loans to business customers issued by their branches and subsidiaries in the early stages of the 2007–2008 crisis?, and 2) did foreign banks react di¤erently from domestic banks to their respective …nancial troubles?

The classic problem with identifying a credit crunch is that …rms’ demand shifts during a credit crunch following the deterioration of …rms’balance sheets. This wouldn’t be an issue if we were studying the cross-border transmission of …nancial distress into an economic area insulated from that distress through all other channels but the bank lending channel. As the sub-prime mortgage crisis was associated since its very beginning with the expectations of a global recession, the measured e¤ect of bank loan supply shocks will likely be contaminated by demand shifts. Some studies that identify demand use the decline in loan applications across di¤erentially a¤ected lenders to argue that there haven’tbeen variations in the decrease in demand across lenders. One problem with that identi…cation approach may be limited data availability on loan applications. However, even when one observes the universe of loan applications, it is still the case that applicant …rms are a trunctated sub-sample of all …rms. Some …rms do not apply because they do not need credit, and some because they are discouraged. This implies a selection process which makes the sample of applicant …rms not perfectly representative of the population. Then it could well be that for banks

2 negatively a¤ected by the crisis, it is the …nancially healthy borrowers that are selecting themselves out of the application process (…rms that do well during a recession), while for other banks, it is the weak …rms that do so, discouraged by news of a contraction in lending. Thus, at di¤erent types of banks, non-applicant …rms may have systematically di¤erent reasons for selecting themselves out of the application process, confounding identi…cation and making it di¢ cult to separate the bank lending from the balance sheet channel.

We overcome this obstacle by employing observable survey information on …rms that choose to select themselves out of the bank credit application process, be it because they were discouraged, or because they do not need credit. Thus we are able to account not just for the decrease in …rms’ demand, but also for the composition of …rms that account for the decrease in demand. While there is already extensive evidence on the real e¤ects of this …nancial crisis3, our paper is the only one we know of which simultaneously 1) studies the international transmission of …nancial distress, 2) accounts for the changes in loan demand, and 3) is able to purge the bank lending channel from the contamination which arises due to the changing composition of …rms demanding bank credit during a recession. As such, our paper adds to a very scarce literature employing data on the selection process involved in the granting of business loans.4

This paper con…rms the hypothesis that the contraction of banks’balance sheets caused by losses on …nancial assets and the deterioration of their equity positions was transmitted cross-border to central and eastern Europe in the relatively early stages of the 2007-2008 crisis. In particular, we

…nd a higher probability of …rms’being credit constrained in localities served by foreign banks whose parents had 1) a low ratio of equity to total assets, 2) a low Tier 1 capital ratio, and 3) high losses on …nancial assets, including ABSs and MBSs.5 The key results hold both when we assume equal access of each …rm to all banks present in the …rm’slocality, or when we weigh access by the branch penetration of each bank. For example, we …nd that in foreign-dominated markets if the

3 De Haas and van Horen (2009), Huang (2009), Ivashina and Scharfstein (2009), Jimenes, Ongena, Peydro, and Saurina (2009), Puri, Rochol, and Ste¤en (2009), Santos (2009), and Tong and Wei (2009) all provide evidence on the credit crunch associated with the 2007-2008 …nancial crisis. 4 The very few studies known to us that do so are Cerqueiro (2009), Chakravarty and Yilmazer (2009), and Ongena and Popov (2009). 5 Initial evidence indicates that high share of mortgage lending to total lending is also associated with transmission of …nancial distress, but due to the many missing data in Bankscope we do not report these …ndings.

3 average Tier 1 capital ratio of the parent of banks present in a particular locality decreases by 2 standard deviations, the probability of …rms in that locality being constrained increases by about

55%. Similarly, we …nd that if the average equity capital ratio by the parent of banks present in a particular locality decreases by two standard deviations, the probability of …rms in that locality being constrained increases by about 30%. It also increases by 32% following a 2-standard deviation deterioration in gains on …nancial assets by banks in that locality. We …nd that foreign banks are more likely to shrink their portfolio in response to …nancial distress, especially low Tier 1 capital ratios, the measure of …nancial distress that is most consistently associated with credit rationing.

Finally, we …nd that …nancial distress is transmitted di¤erently across …rms and industries, in that

…rms that are informationally opaque and …rms with fewer tangible assets su¤er the most.

Our paper relates to a number of studies that have aimed at identifying the transmission of shocks from banks’balance sheets to lending activity in various economic circumstances. The bank lending channel has been studied extensively (e.g., Kashyap and Stein (2000)), and banks have been found to rely heavily on the use of internal capital markets in order to dampen domestic liquidity shocks (e.g., Stein (1997); Houston, James, and Marcus (1997)). The U.S. credit crunch in 1990-92 spawned a large literature that investigated its causes and its e¤ects (e.g., Bernanke and Lown

(1991); Berger and Udell (1994); Peek and Rosengren (1995); Wagster (1996); Hancock and Wilcox

(1998)). Banking crises and liquidity shocks elsewhere in the world similarly generated considerable academic attention (e.g., Woo (1999); Kang and Stulz (2000); Hayashi and Prescott (2002); Khwaja and Mian (2008); Paravisini (2008)). Peek and Rosengren (1997) were one of the …rst to identify the international transmission of …nancial shocks when they investigated how the collapse of asset prices in Japan during the early 1990s a¤ected the operations of Japanese bank subsidiaries abroad.

In particular, they show that the decline in the parents’risk-based capital ratio translated into a signi…cant decline in total loans by the U.S. subsidiaries. Chava and Purnanandam (2009) and

Schnabl (2009) use the exogenous shock provided by the Russian crisis of 1998 to study the e¤ect on lending to U.S. and Peruvian borrowers, respectively. Cetorelli and Goldberg (2008) show that the existence of internal capital markets with foreign bank a¢ liates contributes to an international propagation of domestic liquidity shocks to lending by a¢ liated banks abroad. In the context of

4 the …nancial crisis of 2007-2008, Ivashina and Scharfstein (2009) document that new loans to large borrowers declined by 79% by the end of 2008 relative to the peak of the credit boom (Q2:2007).

They analyze the e¤ect that the failure of Lehman Brothers had on the syndicated loan market to identify the reduction in new lending. Jimenez, Ongena, Peydro, and Saurina (2009) use the universe of bank loans by Spanish banks to identify separately the bank lending channel and the balance sheet channel, and …nd that they dampen each other: more liquid …rms are less vulnerable to the contraction of bank lending, and if banks have ample liquidity, the balance sheet channel partially shuts down. Finally, Puri, Rocholl, and Ste¤en (2009) test the e¤ect of deteriorating balance sheets of U.S. banks on lending to business …rms in Germany. While they account for the shift in …rms’loan demand, they do not account for the variation across lenders in the change in the composition of …rms that select themselves out of the application process.

The paper proceeds as follows. Section 2 presents the data. Section 3 describes the empirical methodology and the identi…cation strategy. Section 4 presents the empirical results. Section 5 concludes with the main …ndings of the paper.

2 Data

The data for our analysis come from three main sources. The core …rm level data come from the 2008 version of the Business Environment and Enterprise Performance Survey (BEEPS), administered jointly by the World Bank and the European Bank for Reconstruction and Development. The survey were carried out between March 10th and April 20th 2008 among 11; 668 …rms from 30 countries in central and eastern Europe and the former Soviet Union. The survey response rate was 36:9%. Surveyees who declined to participate or were unavailable for interviews accounted for

38:3% of the original target group. Firms that were ineligible due to the necessity to ful…ll industry quotas and …rm size quotas accounted for the remainder. We narrowed that sample down to the countries that were most relevant in terms of foreign bank penetration. We complement this data with analogical information on …rms operating in the same countries and localities derived from the 2005 version of the survey. The …nal sample consists of 4; 421 …rms in 14 countries: Albania,

5 Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Macedonia, Montenegro,

Poland, Romania, Slovakia, and Slovenia.

The main purpose of the survey is to obtain information from …rms about their experience with

…nancial and legal constraints, as well as government corruption. In addition, however, BEEPS also included questions about …rm ownership structure, sector of operation, industry structure, export activities, use of external auditing services, subsidies received from central and local governments, etc. Respondent …rms come from 6 di¤erent sectors: construction; manufacturing (11 sub-sectors); transport; wholesale and retail; IT; and hotels and restaurants. The number of …rms covered is roughly proportional to the number of …rms in the country, ranging from 133 in Albania to 709 in

Romania. The survey tried to achieve representativeness in terms of the size of …rms it surveyed: between two thirds and nine tenths of the …rms surveyed are "small" (less than 100 workers) and around 3% of the …rms surveyed are "large" (more than 500 workers).6 The survey also aimed to strike a balance among domestic private, state owned, and foreign owned …rms. Table 2 provides the summary statistics on the number of …rms and their size distribution by country.

For the purpose of estimating the e¤ect of the …nancial crisis on business lending, we focus on the information on credit constraints faced by the …rms in the past …scal year. Question K16 asks: "Did the establishment apply for any loans or lines of credit in …scal year 2007?"7 For …rms that answered

"No" to K16, Question K17 subsequently asks: "What was the main reason the establishment did not apply for any line of credit or loan in …scal year 2007?". For …rms that answered "Yes" to

K16, Question K18a subsequently asks: "In …scal year 2007, did this establishment apply for any new loans or new credit lines that were rejected?". Firms that answered "No need for a loan" to K17 were classi…ed as …rms that do not desire bank credit. Firms that answered "Yes" to

K18a or "Interest rates are not favorable", "Collateral requirements are too high", "Size of loan and maturity are insu¢ cient", or "Did not think it would be approved" to K17 were classi…ed as constrained. The latter classi…cation is in line with the uno¢ cial de…nition by the US Federal

6 See http://www.ebrd.com/country/sector/econo/surveys/beeps.htm for further detailed reports on the representativeness of the survey. 7 Fiscal year 2007 refers to the calender year 2007. However, for tax purposes, in most of the countries in the sample …rms can choose to extend it to March 31, 2008, which is precisely why the Survey was administered in March-April 2008. Given that signs of credit crunch started emerging right after August 9, 2007, the data gives us at least two and at most three quarters of credit crunch potentially experienced by …rms.

6 Reserve of a credit crunch, i.e., a simultaneous increase in the price and decrease in the availability of credit.8 This strategy of grouping …rms that were turned down and …rms that were discouraged from applying is also employed in Cox and Jappelli (1993) and is standard in studies that rely on detailed questionnaires.9 Also, it is crucial given our empirical strategy to separate the …rms that did not apply for credit because they didn’t need it from those that did not apply because they were discouraged. Table 3 presents summary by country of the shares of …rms in need of bank loan and of constrained …rms. Appendix 1 explains the construction of all …rm-level variables in the data.

In addition to the information described above, BEEPS contains information on the locality of the operation of each …rm. A total of 1; 266 localities are present in the data, for an average of

3:5 …rms per locality. That geographic information was then matched with data on bank presence coming from the central banks of the 14 countries involved in the study.10 For the sake of manage- ability, we narrowed our focus to the banks that comprise at least 90% of the banking sector assets in each country. This gives us a range of between 4 banks in Estonia and 9 banks in Bulgaria.

Given this criterion, we performed an internet search of the localities in which each of the banks of interest have branches, as well as the total number of branches in each locality. The search determined that the 1; 266 localities were served by a total of 141 banks. Out of those, 27 are domestic banks, and 114 are branches or subsidiaries of 25 foreign banks. There is considerable variation in foreign bank penetration in the sample: in 2008, foreign ownership of bank sector assets ranges from 28:8% in Slovenia to 98:9% in Slovakia.

Next, we used Bankscope to extract balance sheet information on those 141 bank. We collected data from 2005 to 2008 in order to evaluate how (both current and ex-post) …nancial performance is associated with a potential reduction in credit. We chose our potential explanatory variables in the context of the main issues surrounding the …nancial crisis of 2007-2008. The bursting of

8 The origin can be traced to Bernanke and Lown (1991) who de…ne a credit crunch as a "[...] signi…cant leftward shift in the supply curve for loans, holding constant both the safe real interest rate and the quality of potential borrowers". 9 Using data on central and east European …rms, Brown, Ongena, Popov, and Yesin (2009) show that the share of …rms discouraged from applying is up to twice as large than the share of …rms which applied and had their loan application rejected. 10 The matching was made possible after an extensive research of the web pages of all banks involved. In quite a few cases, information was only available in the respective national language.

7 the housing bubble forced banks to write down several hundred billion dollars in bad loans caused by mortgage delinquencies. At the same time, the stock market capitalization of the major banks declined by more than twice that amount. The total loss on …nancial assets globally is estimated in the neighborhood of trillions of dollars. Central banks around the world pumped hundreds of billions of dollars in short-term liquidity, alongside reducing discount rates at an unprecedented speed, in order to prop up illiquid and likely insolvent banks (Brunnermeier (2009)).

Hence, we focused primarily on banks’pro…t, capital ratios (Tier 1 and total), mortgage lending as a share of the loan portfolio, customer lending as a share of the loan portfolio, problem loans, equity ratios, money market funding, loss on …nancial assets, and loss on available for sale securities.

In the case of foreign ownership, we focused on the …nancial position of the parent bank in order to study, for example, how the investment allocation of UniCredit Group into MBSs and the loss of capital associated with this allocation a¤ects business lending by international branches and subsidiaries of UniCredit. Table 4 summarizes the main variables of interest which were used in the …nal empirical tests. There are apparent cross-country di¤erences - for example, in 2007-2008

Romanian banks had a somewhat low average Tier 1 capital ratio (6:46), close to the 4% regulatory requirement, owing to the relative undercapitalization of their parent foreign banks, while Polish banks had an average Tier 1 capital ratio of 9:9, mostly due to the fact that the largest bank in Poland is the well-capitalized domestic bank PKO Bank Polski. Also, the banks present in

Macedonia incurred almost no losses on …nancial assets in 2007-08, while in 2008 the parents of the banks present in the Czech Republic had an average ratio of gains on …nancial assets to total assets of 0:15. Appendix 2 illustrates the degree of foreign bank penetration in each country in the sample.

Clearly, a group of 22 west European and U.S. banks controls the vast majority of assets in the region. These are Erste Group, Hypo Group, Rei¤eisen, and Volksbank (Austria), Dexia and KBC

(Belgium), Danske Pank (Denmark), Nordea Bank (Finland), Societe Generale (France), Bayerische

Landesbank and Commerzbank (Germany), Alpha Bank, EFG Eurobank, Emporiki Bank, National

Bank of Greece, and Piraeus Bank (Greece), Intessa San Paolo and UniCredit Group (Italy), ING

Bank (Netherlands), Swedbank and Skandinaviska Enskilda Bank (Sweden), and Citibank (U.S.).

8 There is also substantial regional variation in the degree of penetration: for example, the Greek banks operate mostly in south-eastern Europe, the Scandinavian banks in the Baltic countries, and the Austrian banks in central Europe. In addition, there is one domestic "global" bank, the

Hungarian OTP, as well as cross-border penetration by, for example, Parex Group - Latvia and

Snoras Bank - Lithuania.

3 Empirical methodology and identi…cation

3.1 Main empirical model

We start by using the 2008 cross-section data on bank balance sheets, …rm characteristics, and credit constraints to check for a "credit crunch" by estimating the following basic model:

Yijkl = Xijkl + F inancejk + Dk + Dl + "ijkl (1) 1 2 3 4

where Yijkl is a dummy variable equal to 1 if …rm i in city j in country k in industry l is credit constrained in …scal year 2007; Xijkl is a matrix of …rm characteristics; F inancejk is the index of bank health in city j in country k; Dk is a matrix of country dummies; Dl is a matrix of industry dummies; and "ijkl is an idiosyncratic error term. The …rm level characteristics control for observable …rm-level heterogeneity. The two sets of dummy variables control for any unobserved market and industry variation. Essentially, they eliminate the contamination of the estimates by sectoral and macroeconomic circumstances, like growth opportunities, taxes, or unemployment, that are common to all …rms in a particular country.

Next, we pool the 2005 and 2008 samples in order to be able to conduct a proper pre-post analysis using both …rms that were observed in 2007/2008 (the beginning of the …nancial crisis) and in 2004/2005 (the peak of the credit cycle). We estimate the model

Yijkl = Xijkl + F inancejk + Dk + Dl + Dt + "ijkl (2) 1 2 3 4 5

That procedure is analogical to (1), with the exception that we include year …xed e¤ects to

9 account for the change in common macro factors between 2005 and 2008.

The main parameter of interest in both models is 2, which measures the e¤ect of …nancial distress of the banks in each locality on credit access by …rms in that locality. As lower values of

F inance are associated with bigger bank distress, we expect the sign of 2 to be negative. We construct our bank distress index by aggregating balance sheet information from Bankscope after determining which banks were present in that locality, and the original ownership of each bank in that locality. The underlying assumption in the absence of a direct match of each loan to the lending bank and of each rejection to the rejecting bank is that if …rms were granted/denied credit, then it was most likely the result of interaction with banks in the …rms’locality of incorporation.

We use two di¤erent weighting criteria in constructing the index, namely, giving equal weight to each bank in that particular locality, or weighting each bank’s…nancial position by the number of branches it has in the locality.

Here is an example to clarify the above procedure. There are 4 banks in Estonia that hold close to 100% of the banking assets in the country: Swedbank, SEB, Sampo Pank, and Nordea. They are subsidiaries of Swedbank - Sweden, SEB - Sweden, Danske Pank - Denmark, and Nordea - Finland.

In 2008, the 4 parent banks had Tier 1 capital ratios of 8:4, 8:4, 6:9, and 12, respectively. Consider the city Lihula in which only Swedbank has branches. We assign Lihula a Tier 1 capital ratio of

8:4, and then we match the index of …nancial distress in Lihula with all …rms present in that city.

Consider alternatively the city of Kuresaare, in which Swedbank, SEB, and Nordea are present.

They have 2, 1, and 1 branches in that city, respectively. Consequently, in the main analysis, where we assign equal probability of each …rm in that city doing business with each bank present in that city, we assign a Tier 1 capital ratio of 9:6 = 1 8:4 + 1 8:4 + 1 12, which is then matched to 3 3 3 all …rms located in Kuresaare. And in the exercises where we weigh the probability of each …rm doing business with each bank present in Kuresaare by the number of that bank’sbranches in that locality, we assign a Tier 1 capital ratio of 9:3 ( 1 8:4 + 1 8:4 + 1 12). 2 4 4 Now we come to the estimation of the international transmission of …nancial distress. Equations

(1) and (2) simply test for whether banks’ asset and liquidity problems a¤ect business lending, but we also hypothesize that banks with a foreign owner are more likely to do so than domestic

10 banks. For example, if bank-…rm relationships are particularly strong and important, banks may be reluctant to reduce credit to their long-time domestic customers and shift more of the shock to overseas markets (Peek and Rosengren (1997)).

There are two ways in which we address this issue. First, we estimate (1) and (2) on the subsample of localities where more than 67% of the banking assets are controlled by branches or subsidiaries of foreign banks. This gives an answer to the question, do foreign banks transmit …nancial distress. Second, in order to study whether foreign or domestic banks transmit a larger share of their respective …nancial distress, we estimate the following di¤erence-in-di¤erences speci…cation:

Yijkl = Xijkl+ F inancejk FBjk + F inancejk + FBjk + Dk + Dl+ Dt+"ijkl (3) 1 2 3 4 5 6 7

where FBjk is the share of total branches in city j in country k of banks which represent branches or subsidiaries of foreign banks. The primary control group here is all …rms incorporated in locations with little foreign bank penetration. Now 2 measures whether for the same degree of …nancial distress, foreign banks translate more of it into loan application rejections. Consistent with our hypothesis, we expect the sign of 2 to be negative. While in our speci…cations so far we are capable of estimating the e¤ect of …nancial distress net of industry-wide and country-wide recession developments that are common to all …rms in the respective industry (country), they don’t allow us to test whether …nancial distress di¤erentially a¤ects …rms, and our estimates are prone to contamination by location-speci…c unobservables.

Regarding the …rst point, it is generally predicted that informationally opaque …rms and …rms with fewer tangible assets are more likely to be shut out of credit markets (see, for example, Berger, Ofek, and Swary (1996), Beck, Demirgüç-Kunt, and Maksimovic (2005), and Brown, Jappelli, and Pagano

(2009)). Regarding the second one, macroeconomic circumstances like unemployment usually vary at the city level, and so our speci…cation so far will be contaminated by this variation. To address both points, we employ our third and …nal speci…cation

11 Yijkl = Xijkl + F inancejk Zl + Dl + Djk + "ijkl (4) 1 2 3 4

Now the location dummies in Djk absorb the e¤ect of locality-speci…c unobservables. The interaction term containing the industry-level benchmark for information opacity and asset tangibility in Zl allows us to measure whether the potential e¤ect of the credit crunch is indeed strongest for those …rms which theory predicts are most vulnerable to credit market shutdowns.

Finally, we need to emphasize that throughout the paper, it is implicitly assumed that the e¤ect of bank …nancial distress is localized and realized predominately by …rms headquartered in the locality in which the bank has operations. All our empirical speci…cations presume that …rms borrow from banks located near their address of incorporation, which is identical to the approach in, for example, Gormley (2009). In general this is expected to hold as banks tend to derive market power ex ante from geographical proximity (e.g., Degryse and Ongena (2005)). Lending support to that conjecture, empirical work regarding lending relationships in di¤erent countries has demonstrated that the average distance between …rms and banks is usually very small. For example, Petersen and Rajan (2002) …nd that the median distance between a …rm and its main bank in 1993 was only …ve miles (eight kilometers).

3.2 Isolating demand shocks

It is a common challenge of studies that analyze the association between …nancial distress and bank lending to isolate supply shocks satisfactorily. Namely, it is likely that not only does loan demand weaken for all …rms in periods when bank capital declines, but the composition of …rms that demand credit during recessions changes. The solutions to this problem vary in the literature. For example, Peek and Rosengren (1997) bypass this issue by claiming that the identi…cation problem is rather weak in the case of the international transmission of …nancial shocks into a recession-free environment. However, the …nancial crisis of 2007-2008 was followed by one of the deepest global recessions in postwar history, and this recession was already being predicted as soon as the extent of the sub-prime mortgage meltdown became apparent in late summer 2007. Hence, as we observe the

12 …rms in our sample in late 2007 and early 2008, it is conceivable that they were already behaving in a way consistent with a global recession environment. Puri, Rocholl, and Ste¤en (2009) and

Jimenez, Ongena, Peydro, and Saurina (2009) incorporate data on loan applications to account for the explicit weakening of the …rm balance sheet channel. However, this strategy does not account for the changing composition across business lenders of …rms that demand bank credit as these studies do not observe …rms which select themselves out of the loan application process due to 1) weak own demand for loans, or to 2) being discouraged by the deteriorating lending environment.

Not accounting for that estimating the true extent of the transmission of …nancial distress with a bias.

As we explained in Section 2, we eliminate the contamination of the estimates induced by 2) by incorporating data on discouraged …rms in the measure of credit constraint. As for 1), we eliminate the e¤ect of the balance sheet channel by incorporating observable information on …rms which did not apply for bank credit in …scal year 2007 because they did not need one (see Section 2 for the exact de…nition). We apply Heckman’s(1979) selection procedure to eliminate the bias arising from the left-truncation of the sample in that sense. Thus, credit constraint is only observable when a

…rm actually applies for a loan, and the …rm only does so if it needs one, or it’s not discouraged.

Let the dummy variable Q equals 1 if the …rm desires positive bank credit and equals 0 otherwise.

The value of Q is in turn determined by the latent variable:

q = Zijkl + "ijkl (5)

where Zijkl contains …rm and location variables that may e¤ect the …rm’s …xed costs and convenience associated with using bank credit. The variable Q = 1 if q > 0 and Q = 0 otherwise.

2 The error "ijkl is normally distributed with mean 0 and variance . Models (1)-(4) are then (q) (q) updated by adding the term (q) to the RHS, where (q) is the inverse of Mill’s ratio (Heckman (1979)), and where q has been estimated on a set of variables that is larger by at least one variable then the set of variables in (1), (2), (3), and (4), respectively.

13 4 Empirical results

4.1 Bank credit application

Before considering our main empirical model, we …rst consider the bank credit application tests that we use for our Heckman selection correction. Table 5 presents the results from the …rst stage probit regression. The probability of needing bank credit is higher for …rms in more …nancially distressed localities, in the sense of a low bank equity ratio and large losses on …nancial assets.

This implies that not accounting for that selection would bias the estimates of the transmission of

…nancial distress towards zero. If there is such transmission, then by making sure that it is not the case that the …nancially strong …rms are selecting themselves out of the application process in locations where banks are severely distressed, we will be measuring an even stronger e¤ect than when selection is not accounted for.

The need for bank credit increases in the size of the …rm, which is somewhat surprising as one would expect small …rms to have a higher preference for bank credit. However, in a beginning-of-a- recession environment it might be that small …rms are better equipped to …nance investment with cash ‡ows than - potentially - more highly leveraged large …rms. In addition, some of the size e¤ects may be picked by ownership and structural characteristics, as sole properitorships and stand-alone

…rms have a higher demand for loans. The probability of desiring credit is higher for exporters potentially due to their faster expansion, and for audited …rms, which might simply imply that …rms choose to be audited (i.e., they are willing to pay for transparency) when they plan to apply for bank credit.11 It may also be the case that audited …rms have access to …nancial statement lending which may be a cheaper lending technology. In all, these results justify our selection procedure:

…nancial distress not only (potentially) a¤ects business lending, but also the degree to which …rms demand loans. Not accounting for this will introduce bias into the main estimates.

11 The results are broadly consistent with Ongena and Popov (2009) who apply a double selection technique to the BEEPS 2005 sample.

14 4.2 Transmission of …nancial distress

4.2.1 Nonparametric di¤erence-in-di¤erences estimates

Table 6 gives a simple non-parametric illustration of the validity of our empirical strategy. We average the data on rejection rates across localities for 2005 and 2009, and for two distinct criteria: a¤ected vs. non-a¤ected, and foreign-bank dominated vs. domestic bank-dominated. In determining which localities are a¤ected, we look at Tier 1 capital and de…ne "a¤ected" as localities where the average Tier 1 capital ratio of banks present in that locality decreased between 2005 and 2008.

Likewise, we de…ne "foreign-bank dominated" as localities where more than 2/3 of the branches of all banks present in that localities are held by subsidiaries or branches of foreign banks. The table implies that our estimates depend on the di¤erential response over time of foreign vs. domestic banks to their respective …nancial problems. In particular, on average rejection rates went down substantially between 2005 and 2008 in localities populated by banks which were not a¤ected by the crisis, regardless of whether these were foreign or domesetic banks. This in itself may potentially re‡ect a general easing of credit over time or an improvement in the pool of borrowers. However, in localities populated by a¤ected banks, credit tightened between 2005 and 2008. Importantly, while in localities dominated by a¤ected domestic banks rejection rates increased from 27.1% to 30.1%, and the increase is not signi…cant, in localities populated predominantly by a¤ected foreign banks rejection rates went from 30.4% to 42.3%, and that increase is signi…cant at the 5% level.

4.2.2 Cross-section results

Table 7 reports the estimates of the e¤ect of parent banks’…nancial distress on credit constraints faced by local …rms for all …rms present in BEEPS 2008. We report the results of the model in equation (1) alongside the results from the Heckman selection-corrected version of it in order to contrast the two approaches. The three main explanatory variables of interest are: the ratio of equity over total assets; the Tier 1 capital ratio; and the gain on …nancial assets over total assets.12 We …rst report the results from the model in which each bank is given equal weight in

12 The e¤ect of other variables, like the ratio of problem loans to total loans, pro…t, money market funding, and gain on available for sales securities was tested, but the results were insigni…cant.

15 each locality where the bank is present (Panel A). As expected, all else equal, small …rms are more credit constrained, potentially indicating lower ability to tap alternative capital markets, and …rms that export part of their production are less constrained, potentially signalling the willingness of banks to lend to …rms with higher growth prospects.

Turning to the variables of interest, only the Tier 1 capital ratio turns out to have a signi…cant impact on the probability of …rms being constrained in the credit market. The e¤ect has the predicted sign, namely, banks whose parents have lower regulatory capital ratios tend to restrict credit access more. The result continues to hold once we eliminate the e¤ect of the balance sheet channel by integrating out the unobservable information associated with the decision to apply for bank credit or to select oneself out of the application process. Numerically, a 2-standard deviation decrease in average Tier 1 capital ratio for banks in a particular locality increases the probability of …rms in that locality being constrained by about 24%.

When we apply the second weighting criterion in Panel B, namely, weighting the probability of the …rm doing business with each particular bank by the number of branches the bank has in that locality, the e¤ect of the bank’sTier 1 capital ratio becomes larger and more signi…cant. The interpretation of the latter is roughly identical to the one in Panel A, and a 2-standard deviation decrease in average Tier 1 capital ratio for banks in a particular locality increases the probability of …rms in that locality being constrained by about 40%. The sign of the inverse of Mill’s ratio is generally positive, implying that unobservables which increase the probability of needing bank credit, also increase the probability of being constrained in credit markets.

Although some of the e¤ects are only signi…cant at the 10% level (namely, when we weight banks’Tier 1 capital ratios equally), recall that by looking at …scal year 2007, we are capturing only the initial stages of the crisis up to March 31, 2008. In addition to that, our results are contaminated by months of pre-crisis experience before August 2007. In that sense, if there is bias in our estimates, it only goes against …nding any transmission of crisis-related …nancial distress.

16 4.2.3 Pooled 2008 and 2005 data

We next repeat the empirical tests on the sample of …rms that are present both in the 2008 and the

2005 BEEPS, employing equation (2) and the Heckman selection-corrected version of it. This allows us to account for the changing composition of …rms that select themselves out of the applciation process, going from the peak to the trough of the credit cycle. In other words, the information on whether …rms do not apply for credit because they don’tneed it, or because they are discouraged, and how that changes over time, is used eliminate the potential contamination of our estimates by the correlation between credit needs and bank …nancial health. These results are reported in Table

8. The simple correlation between the probabilities of being constrained in 2005 and 2008 is 0:63, implying that variations in past credit constraints explain a large part of the variation in current ones.13 In this speci…cation, only Tier 1 capital seems to matter for credit constraints, once the e¤ect of demand for credit is eliminated (Panel A). Similar to the full sample, once we weigh the probability of …rms doing business with a particular bank by the number of branches the bank has in the locality of the …rm, the e¤ect of low Tier 1 capital ratio on rejection probabilities becomes larger and signi…cant at the 1% level (Panel B). Importantly, we con…rm that not accounting for selection introduces downward bias. This time, once year e¤ects are eliminated, the sign of the inverse of

Mill’s ratio is generally negative, implying that unobservables which increase the probability of needing bank credit, also decrease the probability of being constrained in credit markets.

4.3 International transmission of …nancial distress

The analysis so far tests the international transmission of …nancial distress indirectly, by matching local loan access data to balance sheet data on parent banks. However, as illustrated by Table 4, still relatively large portions of the banking sector are owned by domestic banks. That share in

2008 is 21% for the sample, 24% for Poland, 36% for Hungary and Latvia, and 71% for Slovenia.

In essence, so far we have measured the transmission of distress from the …nancial to the corporate sector regardless of bank ownership. For that reason, we next improve the model by restricting the sample to localilties where more than 2/3 of the assets of present banks are held by foreign

13 In all tables to follow, only coe¢ cients of interest are reported for brevity.

17 banks. That share is calculated individually for each locality by calculating the share of retail branches held by subsidiaries and branches of foreign banks present in that particular locality by the total number of bank branches in that locality. We then look at localities with at least 2/3 foreign presence.

Table 9 reports the estimates of this empirical exercise on the international transmission of

…nancial distress during the 2007-2008 crisis. It turns out that once we restrict our attention to foreign bank-dominated localities, all measures of …nancial distress matter. In other words, higher …nancial distress is associated with lower loan granting probability when …nancial distress is measured as a low equity ratio, a low Tier 1 capital ratio, or high losses on …nancial assets.

Numerically, a 2-standard deviation decrease in equity capital, Tier 1 capital, and gains on …nancial assets is associated with a 30%, 55%, and 32% increase in rejection rates, respectively. This result holds after the correction for the possibility that weak …rms select themselves out of the application process in the case of Tier 1 capital, and only appears after accounting for selection in the case of equity capital for the 2008 sub-sample.

In unreported tests, we check if the geographic origin of the main foreign owner matters. This is interesting question given that we study a heterogeneous group of foreign banks, that is, there are countries in the sample dominated by Greek banks vs. countries dominated by Austrian and

Belgian banks vs. countries dominated by Skandinavian banks. The di¤erent dimension of the domestic recession in these countries may have led some of them to cut supply abroad more. For example, it is possible to think of the exposure of Dutch and Spanish banks to domestic mortgages and hypothesize that these banks would want to keep good relations with their domestic depositors by transmitting a smaller portion of an identical …nancial shock domestically and, by extension, a larger portion internationally. However, we …nd that this is not the case.

4.4 Transmission of …nancial distress: foreign vs. domestic

An important question that arises given the evidence so far is, do foreign or domestic banks transmit a larger portion of an identical …nancial shock. Table 10 reports the estimates from the di¤erence- in-di¤erences regression where we compare the transmission of …nancial distress by domestic and

18 foreign banks, that is, equation (3). Tellingly, whenever signi…cant, the interaction e¤ect implies that foreign banks react to the same shock to balance sheets by shrinking their portfolio more than domestic banks. This is observed in the case of shocks to Tier 1 capital and to equity capital. The one exception is gains on …nancial assets, where we …nd weak evidence to the opposite e¤ect. The positive sign on the coe¢ cient implies that for the same level of bank distress so de…ned, …rms in localities dominated by foreign banks face a lower probability of being credit constrained. This one result implies that domestic banks indeed shrink their loan portfolios more than foreign owned banks in response to asset losses. However, it is only signi…cant in one case, and only at the 10% level. Consequently, we can conclude with a fair degree of statistical certainty that foreign banks transmit more of an identical …nancial shock than domestic banks.

Apart from a parametric con…rmation of the non-parametric observation in Table 6, this result o¤ers important insights into the role of foreign banks in emerging markets. In general, the e¤ect of foreign banks on business lending in the literature is ambiguous. A large literature has found that foreign bank presence is associated with higher access to loans (Clarke, Cull, and Peria (2006)), higher …rm-level sales (Giannetti and Ongena (2009)), and lower loan rates and higher …rm leverage

(Ongena and Popov (2009)). On the other hand, Berger, Klapper, and Udell (2001), Mian (2006), and Gormley (2009) show that foreign banks tend to …nance only larger, established, and more pro…table …rms. Our paper complements that picture by providing evidence that while at the peak of the credit cycle lending by foreign banks is indistinguishable from lending by domestic banks in terms of acceptance rates, foreign banks do tend to shrink their loan portfolio following a capital crunch, even after controlling for the degree of …nancial distress.

4.5 Transmission of …nancial distress: di¤erential e¤ects

Finally, we ask which …rms are most a¤ected from the transmission of …nancial distress. There are clear arguments in the literature on which …rms and industries should be most a¤ected by credit rationing. Information asymmetries and the tangibility of the …rm’s assets, for example, are expected to play an important role in explaining di¤erences in credit availability across …rms.

Informationally opaque …rms tend to su¤er more from credit rationing, especially when foreign bank

19 lending is involved (Berger, Klapper, and Udell (2001)). Firms are expected to be informationally opaque when they come from an industry where banks …nd it di¢ cult to evaluate projects and price risk properly. Berger, Klapper, and Udell (2001) also show that …rm size is not a good proxy for informational opacity. However, we know that banks do less business with informationally opaque

…rms, and based on that, we hypothesize that …rms in industries which use a lot of external …nance in mature …nancial markets are less informationally opaque than …rms which use little external

…nance. Regarding asset tangibility, Berger, Ofek, and Swary (1996) show that …rms with less tangible assets are more likely to lose access to credit when banks reprice risk. The rationale is that lenders rely more on collateral when making lending decision rather than investing in costly screening technologies, and this problem will tend to be exacerbated in an environment where risk is suddenly priced higher.

We proceed by collecting data on mature U.S. …rms and using it to construct industry benchmarks for informational opacity and asset tangibility. The rationale for doing so goes back to Rajan and Zingales (1998) who argued that the actual capital structure of small …rms is a function of

…nancial constraints, while the capital structure of large mature …rms is more representative of the cross-industry variations in the scale of projects, gestation period, the ratio of hard vs. soft information, the ratio of tangible vs. intangible assets, follow-up investments, etc. In addition, doing so for large U.S. …rms makes sure that what is taken as a "natural" industry feature is not contaminated by shallow …nancial markets.

Following Rajan and Zingales’s (1998) original approach, we proceed by taking all Compustat

…rms between 1990 and 2000. We …rst exclude all …rms that are young in the sense that they have gone public only recently (in the last 10 years) to make sure that we are not capturing the excessive appetite for funds exhibited during the early life of a public …rm. For each …rm, we sum across all years total capital expenditures minus cash ‡ow, normalized by total sales, and take industry-median values. This is our industry benchmark "dependence on external …nance". Next, we sum across all years each …rm’s ratio of research and development expenses over sales. We take the median industry value of that ratio and this value constitutes our industry benchmark for

"R&D intensity". Finally, we sum across all years each …rm’sratio of total physical capital used in

20 production over the number of employees. The industry median value of that variable constitutes our industry benchmark for "Capital intensity". For each benchmark, we have an 18-industry variation.

Table 11 reports the estimates of equation (4) where each measure of …nancial distress has been interacted with a dummy variable equal to 1 if the industry is in the bottom 50% of the distribution of "External …nancial dependence" (informationally opaque), in the top 50% of the distribution of

"R&D intensity" (low asset tangibility), and in the bottom 50% of "Capital intensity" (low asset tangibility). We only forcus on …nancial distress as measured by low Tier 1 capital ratios, as this is the one measure that is most consistently associated with higher loan rejection rates in the analysis so far. Importantly, this speci…cation gives us interaction at the city and industry level, and thus we can include city dummies in the regression. The direct e¤ect of …nancial distress is now fully absorbed by these dummies, along with any unobservable variation in macroeconomic conditions at the location level. The e¤ect of the natural industry benchmarks is absorbed by the industry dummies.

The results con…rm the intuition: …rms tend to su¤er more from the transmission of …nancial distress when they are informationally opaque or don’thave enough assets to pledge as collateral.

Numerically, the same Tier 1 capital ratio is associated with a 6:0% higher probability of loan rejection for …rms in industries with low dependence on external …nance; with a 7:8% higher probability of loan rejection for …rms in industries with high R&D intensity; and with a 7:1% higher probability of loan rejection for …rms in industries with low per-worker capital.

5 Conclusion

The …nancial crisis of 2007-2008, which started with the meltdown of sub-prime mortgages and securitized products and which has been characterized by severe losses and depletion of bank capital, has spurred unprecedented government recapitalization programs and liquidity injections by central banks. Since the inception of the crisis, it was feared that this depletion of capital may result in a severe credit crunch, especially to the corporate sector in countries populated by the

21 hardest hit banks. Because the European economy is heavily bank-dependent and SMEs - usually the most vulnerable to a credit crunch due to their opacity - comprise up to 99% of the corporate sector, it was feared that European …rms would be particularly heavily hit, despite the fact that the causal factors of the credit crunch originated elsewhere.

In this paper, we investigate empirically the international transmission of …nancial distress, from the loss in value of …nancial assets to the balance sheets of big European and U.S. banks to business lending in their foreign markets - speci…cally, central and eastern Europe. Several current unpublished studies have documented a credit crunch associated with weakened capital positions, however, ours is the …rst one to simultaneously 1) demonstrate the cross-border dimensions of this phenomenon, and 2) eliminate the contamination of the lending channel by selection bias resulting from the changing composition of …rms’demand for credit during recessions.

We …nd that di¤erent types of …nancial distress at western European and U.S. parent banks are associated with a signi…cant impact on business lending to central and eastern European …rms.

While we do not observe an actual match between a bank and a …rm, we match …rms and banks by the locality of their respective operation. We …nd that as early as late 2007/early 2008, …rms reported higher credit constraints in localities populated by branches or subsidiaries of banks who in 2007 and 2008 had low equity capital, low Tier 1 capital ratios, and had recorded severe losses on …nancial assets. Importantly, we …nd that this e¤ect is stronger for localities predominantly populated by foreign banks. We also …nd that informationally opaque …rms and …rms with fewer tangible assets were di¤erentially more a¤ected by this capital crunch. These results hold when we eliminate the e¤ect of demand shifts in response to weakening …rm balance sheets, as well as the bias resulting from the systematic selection of …rms out of the application process. Our evidence implies that all else equal, …rms in countries like the Czech Republic and Romania, where major portions of the banking market are held by the relatively undercapitalized Erste Group and UniCredit Group, were 1) more credit constrained than …rms in countries like Hungary and Poland, where the largest banks are the domestically-owned and well capitalized OTP and PKO, respectively, but also 2) more credit constrained than …rms in Slovenia, served by equally undercapitalized domestic banks.

This is direct evidence of the global transmission of …nancial distress in the relatively early stages

22 of the 2007-2008 …nancial crisis, in a way unrelated to the demand for loans in local markets.

The …nancial crisis of 2007-2008 has …nally laid to rest the idea that the e¤ect of large …nancial shocks can be con…ned locally. We have shown how the collapse of housing values in the U.S. has a¤ected the …nancing conditions of Slovak …rms through the deteriorating portfolios of Austrian,

Belgian, and Italian banks, loaded with assets backed by those mortgages, and operating in Slovakia through their subsidiaries. While the credit crunch only started in the third quarter of 2007, banks kept tightening credit standards until as late as the fourth quarter of 200814, and most likely long after that. Thus, despite the coordinated actions of various national and supranational authorities, which kept the global …nancial system from collapsing after the fall of Lehman Brothers in September 2008, it is likely that the losses that the …nancial system endured have induced, and will continue to induce, a much larger impact on the real sector than the one estimated in this paper. The true extent of the credit crunch will only become clear with the availability of new, more comprehensive data.

14 See ECB (2008) for details.

23 References

[1] Beck, T., Demirgüç-Kunt, A., and V. Maksimovic, 2005. Financial and legal constraints to

growth: Does …rm size matter? Journal of Finance 60, 137-177

[2] Berger, A., Klapper, L., and G. Udell, 2001. The ability of banks to lend to informationally

opaque small businesses. Journal of Banking and Finance, 25(12), 2127-2167.

[3] Berger, A., and G. Udell, 1994. Did risk-based capital allocate bank credit and cause a "credit

crunch" in the United States? Journal of Money, Credit, and Banking 26, 585-628.

[4] Berger, P., Ofek, E., and I. Swary, 1996. Investor valuation of the abandonment option. Journal

of Financial Economics, 42(2), 257-287.

[5] Bernanke, B., and C. Lown, 1991. The credit crunch. Brookings Papers on Economics Activity

2, 205-248.

[6] Brown, M., Jappelli, T., and M. Pagano, 2009. Information sharing and credit: Firm-level

evidence from transition countries. Journal of Financial Intermediation 18, 151-172.

[7] Brown, M., Ongena, S., Popov, A., and P. Yesin, 2009. Who needs credit and who gets credit

in emerging Europe? European Central Bank mimeo.

[8] Brunnermeier, M., 2009. Deciphering the liquidity and credit crunch 2007-2008. Journal of

Economic Perspectives 23, 77-100.

[9] Cerqueiro, G., 2009. Bank concentration, credit quality and loan rates. CentER mimeo.

[10] Cetorelli, N., and L. Goldberg, 2008. Banking globalization, monetary transmission, and the

lending channel. NBER Working Paper No. W14101.

[11] Chakravarty, S., and T. Yilmazer, 2009. A multistage model of loans and the role of relation-

ships. Financial Management (forthcoming).

[12] Chava, S., and A. Purnanandam, 2009. The e¤ect of a banking crisis on bank-dependent

borrowers. Journal of Financial Economics (forthcoming).

24 [13] Clarke, G., Cull, R., and M. Peria, 2006. Foreign banks participation and access to credit

across …rms in developing countries. Journal of Comparative Economics 34, 774-795.

[14] Cox, D., and T. Jappelli, 1993. The e¤ect of borrowing constraints on consumer liabilities.

Journal of Money, Banking and Credit 25, 197-213.

[15] De Haas, R., and N. van Horen, 2009. The crisis as a wake-up call: Do banks tighten lending

standards during a …nancial crisis? European Bank for Reconstruction and Development

mimeo.

[16] Degryse, H., and S. Ongena, 2005. Distance, lending relationships, and competition. Journal

of Finance 51, 231-266.

[17] EBRD, 2007. "Credit Crunch". Interview with Erik Berglof, Chief

economist of the European Bank for Reconstruction and Development.

http://www.ebrd.com/new/stories/2007/071003.htm

[18] ECB, 2008. "Financial Stability Review." December 2008.

[19] Giannetti, M., and S. Ongena, 2009. Financial integration and …rm performance: Evidence

from foreign bank entry in emerging markets. Review of Finance 13, 181-223.

[20] Gormley, T., 2009. The impact of foreign bank entry in emerging markets: Evidence from

India. Journal of Financial Intermediation (forthcoming).

[21] Gorton, G., 2009. Slapped in the face by the invisible hand: Banking and the panic of 2007.

Yale School of Management mimeo.

[22] Hancock, D. and J.A. Wilcox, 1998. The "credit crunch" and the availability of credit to small

business. Journal of Banking and Finance 22, 983-1014.

[23] Hayashi, F., and E. Prescott, 2002. The 1990s in Japan: A lost decade. Review of Economic

Dynamics 5, 206-235.

[24] Heckman, J., 1979. Sample selection bias as speci…cation error. Econometrica 47, 153-161.

25 [25] Houston, J., James, C., and D. Marcus, 1997. Capital market frictions and the role of internal

capital markets in banking. Journal of Financial Economics 46, 135-164.

[26] Ivashina, V., and D. Scharfstein, 2009. Bank lending during the …nancial crisis of 2008. Harvard

Business School mimeo.

[27] Jimenez, G., Ongena, S., Peydro, J., and J. Saurina, 2009. Monetary policy and credit crunch:

Identifying simultaneously the bank lending and balance sheet channels. European Central

Bank mimeo.

[28] Kang, J., and R. Stulz, 2000. Do banking shocks a¤ect borrower …rm performance? An analysis

of the Japanese experience. Journal of Business 73, 1-23.

[29] Kashyap, A., and J. Stein, 2000. What do a million observations on banks say about the

transmission of monetary policy? American Economic Review 90, 407-428.

[30] Khwaja, A., and A. Mian, 2008. Tracing the impact of bank liquidity shocks: Evidence from

an emerging market. American Economic Review 98, 1413–1442.

[31] Mian, A., 2006. Distance constraints: The limits of foreign lending to poor economies. Journal

of Finance 61, 1465-1505.

[32] Ongena, S., and A. Popov, 2009. Interbank market integration, loan rates, and …rm leverage.

European Central Bank mimeo.

[33] Paravisini, D., 2008. Local bank …nancial constraints and …rm access to external …nance.

Journal of Finance 63, 2161-2193.

[34] Peek, J., and E. Rosengren, 1995. The capital crunch: Neither a borrower nor a lender be.

Journal of Money, Credit and Banking 27, 625-638.

[35] Peek, , J., and E. Rosengren, 1997. The international transmission of …nancial shocks: The

case of Japan. American Economic Review 87, 495-505.

26 [36] Petersen, M., and R. Rajan, 2002. Does distance still matter? The information revolution in

small business lending. Journal of Finance 57, 2533-2570.

[37] Puri, M., Rocholl, J., and S. Ste¤en, 2009. The impact of the U.S. …nancial crisis on global

retail lending. Fuqua School of Business Duke University mimeo.

[38] Rajan, R., and L. Zingales, 1998. Financial dependence and growth. American Economic

Review, 88(3), 559-586.

[39] Santos, J., 2009. Bank corporate loan pricing following the sub-prime crisis. Federal Reserve

Bank of New York mimeo.

[40] Schnabl, P., 2009. Financial globalization and the transmission of credit supply shocks: Evi-

dence from an emerging market. New York University Working Paper No. FIN-08-008

[41] Stein, J.C., 1997. Internal capital markets and the competition for corporate resources. Journal

of Finance 52, 111-134.

[42] Tong, H., and S.-J. Wei, 2009. The misfortune of non-…nancial …rms in a …nancial crisis:

Disentangling …nance and demand shocks. CEPR discussion paper 7208.

[43] Wagster, J. D., 1996. Impact of the 1988 Basle Accord on international banks. Journal of

Finance 51, 1321-1346.

[44] Woo, D. 2003. In search of capital crunch: Supply factors behind the credit slowdown in Japan.

Journal of Money, Credit and Banking 35, 1019-1038.

27 Figure 1. Origin and target countries in the data

The map shows the cross-border dimension of the underlying data. Countries in dark color (Austria, Belgium, Denmark, Finland, France, Germany, Greece, Netherlands, and Sweden) are those in which the parent banks in the dataset are incorporated. Countries in light color (Albania, Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia, Lithiania, Macedonia, Montenegro, Poland, Romania, Slovakia, and Slovenia) are those where the firms in the dataset are incorporated.

28 Table 1. Timeline of events during the 2007-2008 crisis concerning banks and countries in the data

Timeline Country Event

Germany Bayerische LandesBank is one of three LandesBanken Aug. 2007 – Aug. 2008 to receive capital injections, credit lines, and asset- backed securities loss guranatees. Sept. 2008 France The government recapitalizes Dexia. U.S. Emergency Economic Stabilization Act, containing a commitment for up to 700 bln. USD to purchase bad assets from banks. Italy The parliament approves a law granting the government the possibility to recapitalize distressed banks. Netherlands The government announces that public fudns can be used for bank recapitalization, of which 20 bln. EUR are immediately available. Oct. 2008 France The Government approves 320 bln. EUR to provide loans to banks and other financial firms, including a 40 billion euro recapitalization plan. Sweden The government announces that it will guarantee up to 1.5 trillion SEK in new debt issues, and a 15 billion SEK stabilization fund. Germany The government announces a 400 billion EUR plan to guarantee bank financing, including a 70 billion EUR recapitalization fund. US The Treasury subscribes 20 bln. USD preferred shares at Citigroup and ring-fences its troubled assets worth up to 300 billion USD. Nov. 2008 Italy The government approves a law to inject capital into sound banks. Germany Bayerische LandesBank receives 7 billion EUR of capital from the Bavarian state. Dec. 2008 Germany The Finance ministry provides Bayerische LandesBank with 15 billion EUR . Germany The Finance ministry provides Commerzank with a 8.2 billion EUR loan, and buys 1.8 trillion EUR worth of equity. Jan. 2009 France The government implements a second round of bank recapitalization for 10.5 billion EUR. Netherlands The Dutch government provides a cack-up facility to back up the risks of ING’s securitized mortgage portfolio worth 35.1 billion EUR.

29 Table 2. Summary statistics: Firm characteristics

Public Private Sole country # firms Small firm Big firm company company proprietorship Privatized Subsidized Exporter Competition Audited Albania 133 0.90 0.04 0.00 0.33 0.63 0.08 0.03 0.27 0.65 0.63 Bulgaria 457 0.84 0.02 0.05 0.47 0.43 0.14 0.07 0.25 0.63 0.43 Croatia 208 0.81 0.04 0.06 0.51 0.38 0.23 0.20 0.36 0.85 0.47 Czech Republic 316 0.72 0.03 0.07 0.61 0.29 0.09 0.22 0.46 0.82 0.54 Estonia 402 0.78 0.03 0.16 0.62 0.19 0.10 0.16 0.34 0.75 0.80 Hungary 406 0.73 0.06 0.01 0.65 0.31 0.15 0.22 0.34 0.87 0.75 Latvia 385 0.70 0.04 0.02 0.69 0.23 0.14 0.12 0.34 0.80 0.70 Lithuania 386 0.75 0.03 0.02 0.71 0.22 0.15 0.17 0.37 0.78 0.39 Macedonia 477 0.80 0.03 0.06 0.56 0.23 0.17 0.04 0.41 0.85 0.59 Montenegro 134 0.85 0.01 0.04 0.27 0.69 0.12 0.03 0.15 0.68 0.53 Poland 621 0.79 0.02 0.05 0.24 0.67 0.12 0.13 0.28 0.81 0.37 Romania 709 0.70 0.03 0.06 0.66 0.22 0.14 0.10 0.18 0.71 0.38 Slovakia 349 0.70 0.06 0.05 0.18 0.64 0.13 0.15 0.34 0.79 0.57 Slovenia 397 0.71 0.05 0.10 0.62 0.25 0.23 0.24 0.61 0.80 0.44 Total 5,380 0.76 0.03 0.06 0.53 0.36 0.14 0.14 0.33 0.77 0.53

Note: The table presents statistics on the number of firms and the share of firms by size, ownership, privatization history, subsidies from central and local governments, access to foreign product markets, degree of competition, and access to international auditing, by country. See Appendix 1 for exact definitions. Source: BEEPS (2008 & 2005).

30 Table 3. Summary statistics: Credit demand and access

2008 2005 Country Need loan Constrained Need loan Constrained Albania 0.29 0.47 0.74 0.46 Bulgaria 0.58 0.52 0.65 0.45 Croatia 0.59 0.42 0.77 0.19 Czech Republic 0.53 0.32 0.78 0.38 Estonia 0.54 0.27 0.58 0.47 Hungary 0.41 0.31 0.88 0.42 Latvia 0.59 0.48 0.74 0.58 Lithuania 0.60 0.23 0.70 0.61 Macedonia 0.59 0.50 0.77 0.82 Montenegro 0.78 0.48 0.80 0.45 Poland 0.53 0.41 0.66 0.59 Romania 0.61 0.33 0.75 0.47 Slovakia 0.53 0.40 0.60 0.38 Slovenia 0.64 0.15 0.77 0.12 Total 0.57 0.37 0.72 0.46

Note: The table presents statistics on the share of firms who declare bank loans desirable, and the share of firms out of those that need a lon that have been formally rejected or did not apply because they found access to finance too difficult, by country. The data are for the fiscal year 2007 (until March 31, 2008). See Appendix 1 for exact definitions. Source: BEEPS (2008).

31 Table 4. Bank ownership balance sheet data

2005 2008 2005 2008 2005 2008 2005 2008 Country % foreign owned bank assets Equity/assets Tier 1 capital ratio Gain on financial assets Albania 0.92 0.94 0.068 0.051 9.01 7.01 0.02 -0.04 Bulgaria 0.75 0.82 0.065 0.061 9.65 7.43 0.06 -0.03 Croatia 0.91 0.90 0.073 0.063 7.84 7.03 0.06 -0.02 Czech Republic 0.82 0.86 0.049 0.044 8.74 7.26 0.09 -0.15 Estonia 0.99 0.99 0.049 0.038 9.39 7.57 0.05 -0.03 Hungary 0.83 0.64 0.071 0.058 8.51 7.38 0.01 -0.06 Latvia 0.58 0.64 0.084 0.062 9.31 9.76 0.02 -0.06 Lithuania 0.92 0.92 0.061 0.053 8.59 7.68 0.03 -0.04 Macedonia 0.51 0.86 0.069 0.065 9.77 7.60 0.06 -0.01 Montenegro 0.88 0.79 0.092 0.091 9.13 8.16 0.02 -0.03 Poland 0.74 0.76 0.066 0.065 9.41 9.91 0.02 -0.09 Romania 0.59 0.87 0.064 0.053 8.44 6.46 0.04 -0.05 Slovakia 0.97 0.99 0.059 0.054 8.60 7.63 0.01 -0.06 Slovenia 0.23 0.29 0.062 0.055 8.08 6.81 0.07 -0.06

Note: The table reports summary statistics on the share of the domestic banking system owned by branches and subsidiaries of foreign banks, of the average ratio of equity financing to total bank assets, of the average Tier 1 capital ratio, and of average gains on financial assets by the parent of the banks operating in each country, by country. The data are averaged for 2007-2008. See Appendix 1 for exact definitions. Source: EBRD Transition Report (2008) and Bankscope (2007 and 2008).

32 Table 5. Probability of desiring bank credit

Finance = Finance = Finance = Equity/assets Tier 1 capital ratio Gains on fin assets Finance -0.056 0.039 -0.023 (0.031)* (0.063) (0.013)* Small firm -0.245 -0.169 -0.169 (0.053)*** (0.060)*** (0.060)*** Big firm 0.082 -0.009 -0.010 (0.117) (0.137) (0.137) Public company -0.008 -0.054 -0.051 (0.124) (0.146) (0.146) Private company 0.151 0.137 0.141 (0.092) (0.114) (0.114) Sole proprietorship 0.200 0.223 0.227 (0.093)** (0.119)* (0.119)* Privatized 0.047 0.094 0.095 (0.061) (0.071) (0.071) Exporter 0.149 0.159 0.1159 (0.043)*** (0.054)*** (0.054)*** Competition 0.089 0.088 0.088 (0.051)* (0.051)* (0.051)* Audited 0.086 1.141 0.140 (0.045)* (0.050)*** (0.049)** Subsidized 0.281 0.281 0.281 (0.069)*** (0.069)*** (0.069)*** Stand-alone firm 0.271 0.271 0.271 (0.075)*** (0.075)*** (0.075)*** Firm age 0.001 0.001 0.001 (0.002) (0.002) (0.002) Manager experience 0.001 0.001 0.001 (0.002) (0.002) (0.002) Country fixed effects Yes Industry fixed effects Yes Observations 4,709 4,709 4,813 Pseudo R-squared 0.06 0.06 0.07

Note: The dependent variable is a dummy variable equal to 1 if the firm desires bank credit. ‘Finance’ is one of the three financial variables from Table 4. Ommited category in firm size is ‘Medium firm’. All regressions include country, industry, and year fixed effects. *** indicates significance at the 1% level, ** at the 5% level, and * at the 10% level. See Appendix 1 for exact definitions. Source: BEEPS (2005 and 2008) and Bankscope (2005 and 2008).

Table 6. Affected vs. non-affected banks: rejection rates

Foreign banks-dominated Affected localities Non-affected localities 2005 0.304 0.357 2008 0.423 0.143 Difference -0.119** 0.214***

Domestic bank-dominated Affected localities Non-affected localities 2005 0.271 0.385 2008 0.301 0.150 Difference -0.030 0.235***

Note: The table presents acceptance rates aggregated across localities. ‘Affected’ are localities where the average Tier 1 capital ratio of banks present in that locality went down between 2005 and 2008. ‘Foreign banks-dominated’ are localities where more than 2/3 of the branches are held by subsidiaries or branches of foreign banks. The statistical significance of the difference-in-differences estimate from a two-sided Mann- Whitney test can be found next to the difference, where *** indicates significance at the 1% level, ** at the 5% level, and * at the 10% level. Source: BEEPS (2005 and 2008) and Bankscope (2005-2008).

Table 7. Probability of being constrained (2008 sample)

Panel A. Equally weighted bank data for each locality

Finance = Finance = Finance = Equity/assets Tier 1 capital Gains on fin assets Finance 0.018 0.028 -0.219 -0.237 0.093 0.325 (0.075) (0.076) (0.132)* (0.116)** (0.651) (0.699) Small firm 0.396 0.560 0.394 0.556 0.395 0.543 (0.084)*** (0.110)*** (0.084)*** (0.110)*** (0.084)*** (0.109)*** Big firm -0.188 -0.141 -0.186 -0.138 -0.187 -0.132 (0.186) (0.194) (0.186) (0.195) (0.186) (0.194) Public company 0.324 0.315 0.323 0.314 0.324 -0.312 (0.216) (0.226) (0.216) (0.226) (0.216) (0.226) Private company -0.103 -0.127 -0.106 -0.129 -0.102 -0.118 (0.171) (0.183) (0.171) (0.183) (0.171) (0.182) Sole proprietorship -0.065 -0.017 0.064 -0.015 0.065 -0.001 (0.177) (0.196) (0.177) (0.196) (0.177) (0.195) Privatized -0.031 -0.034 -0.029 -0.030 -0.031 -0.024 (0.099) (0.107) (0.099) (0.107) (0.099) (0.106) Exporter -0.211 -0.252 -0.214 -0.254 -0.211 -0.242 (0.075)*** (0.087)*** (0.075)*** (0.087)*** (0.075)*** (0.086)*** Audited -0.285 -0.354 -0.283 -0.350 -0.285 -0.343 (0.070)*** (0.085)*** (0.070)*** (0.085)** (0.070)*** (0.085)*** Subsidized -0.072 -0.269 -0.074 -0.268 -0.073 -0.243 (0.090) (0.142)* (0.090) (0.142)* (0.090) (0.140)* Stand-alone firm -0.023 -0.139 -0.021 -0.137 -0.023 -0.121 (0.107) (0.132) (0.108) (0.132) (0.108) (0.131) Inverse Mill's ratio 0.301 0.294 0.247 (0.209) (0.209) (0.202) Country fixed effects Yes Industry fixed effects Yes Observations 2,082 2,005 2,082 2,005 2,082 2,005 Pseudo R-squared 0.12 0.13 0.12 0.13 0.12 0.13

Table 7. Probability of being constrained (2008 sample)

Panel B. Branch-weighted bank data for each locality

Finance = Finance = Finance = Equity/assets Tier 1 capital Gains on fin assets Finance -0.068 -0.070 -0.338 -0.408 0.284 0.204 (0.069) (0.071) (0.124)*** (0.131)*** (0.573) (0.585) Small firm 0.394 0.560 0.393 0.557 0.395 0.571 (0.084)*** (0.110)** (0.084)*** (0.110)*** (0.084)*** (0.110)*** Big firm -0.181 -0.135 -0.189 -0.143 -0.188 -0.142 (0.186) (0.194) (0.186) (0.195) (0.186) (0.194) Public company 0.322 0.312 0.318 0.303 0.325 0.317 (0.216) (0.226) (0.216) (0.227) (0.216) (0.226) Private company -0.105 -0.133 -0.104 -0.133 -0.101 -0.133 (0.171) (0.183) (0.171) (0.183) (0.171) (0.182) Sole proprietorship 0.067 -0.018 0.074 0.008 0.067 -0.027 (0.177) (0.196) (0.177) (0.196) (0.177) (0.196) Privatized -0.029 -0.031 -0.021 -0.020 -0.028 -0.035 (0.099) (0.107) (0.096) (0.107) (0.099) (0.107) Exporter -0.210 -0.253 -0.214 -0.256 -0.211 -0.259 (0.075)*** (0.086)*** (0.075)*** (0.087)*** (0.075)*** (0.086)*** Audited -0.282 -0.352 -0.281 -0.351 -0.284 -0.362 (0.070) (0.085)*** (0.070)*** (0.085)*** (0.070)*** (0.085)*** Subsidized -0.069 -0.269 -0.071 -0.270 -0.073 -0.290 (0.090) (0.141)* (0.090) (0.141)* (0.090) (0.141)** Stand-alone firm -0.020 -0.138 -0.023 -0.141 -0.025 -0.154 (0.108) (0.131) (0.108) (0.103) (0.108) (0.132) Inverse Mill's ratio 0.307 0.302 0.340 (0.207) (0.209) (0.207)* Country fixed effects Yes Industry fixed effects Yes Observations 2,082 2,005 2,082 2,005 2,082 2,005 Pseudo R-squared 0.12 0.13 0.12 0.13 0.12 0.13

Table 8. Probability of being constrained (pooled 2005 and 2008 samples)

Panel A. Equally weighted bank data for each locality Finance = Finance = Finance = Equity/assets Tier 1 capital Gains on fin assets Finance 0.061 0.056 -0.153 -0.175 0.010 -0.001 (0.073) (0.074) (0.110) (0.105)* (0.022) (0.022) Inverse Mill's ratio -0.193 -0.196 -0.190 (0.109)* (0.109)* (0.090)** Country fixed effects Yes Industry fixed effects Yes Year fixed effects Yes Observations 2,398 2,365 2,398 2,365 2,414 2,481 Pseudo R-squared 0.14 0.15 0.14 0.15 0.14 0.15 Panel B. Branch-weighted bank data for each locality Finance = Finance = Finance = Equity/assets Tier 1 capital Gains on fin assets Finance -0.042 -0.041 -0.355 -0.392 0.012 0.011 (0.068) (0.069) (0.121)*** (0.125)*** (0.019) (0.020) Inverse Mill's ratio -0.200 -0.208 -0.194 (0.109)* (0.109)** (0.091)** Country fixed effects Yes Industry fixed effects Yes Year fixed effects Yes Observations 2,398 2,365 2,398 2,365 2,447 2,414 Pseudo R-squared 0.14 0.15 0.14 0.15 0.14 0.15

Table 9. Probability of being constrained, foreign banks dominated localities

Panel A. 2008 sample Finance = Finance = Finance = Equity/assets Tier 1 capital Gains on fin assets Finance 0.137 -0.309 -0.849 -0.899 0.006 0.926 (0.216) (0.179)* (0.435)** (0.346)*** (0.007) (1.285) Inverse Mill's ratio 0.456 0.703 -0.093 0.707 -0.063 0.732 (0.303) (0.354)** (0.363) (0.357)** (0.366) (0.353)** Country fixed effects Yes Industry fixed effects Yes Observations 1,568 1,167 1,244 1,167 1,244 1,167 Pseudo R-squared 0.13 0.15 0.09 0.16 0.09 0.15 Panel B. Pooled 2008 and 2005 samples Finance = Finance = Finance = Equity/assets Tier 1 capital Gains on fin assets Finance -0.397 -0.286 -0.849 -0.812 -0.369 0.065 (0.223)* (0.169)* (0.503)* (0.294)*** (0.221)* (0.055) Inverse Mill's ratio 0.011 0.007 0.005 -0.002 -0.029 -0.032 (0.144) (0.143) (0.144) (0.144) (0.110) (0.112) Country fixed effects Yes Industry fixed effects Yes Year fixed effects Observations 1,383 1,383 1,383 1,383 1,414 1,414 Pseudo R-squared 0.16 0.16 0.16 0.17 0.16 0.16

Table 10. Probability of being constrained: foreign vs. domestic banks

Panel A. 2008 sample Finance = Finance = Finance = Equity/assets Tier 1 capital Gains on fin assets Finance*Foreign -0.451 -0.150 -0.792 -0.086 0.329 0.002 (0.494) (0.210) (0.376)** (0.180) (0.184)* (0.182) Finance 0.060 -0.025 -0.260 -0.390 -0.177 0.252 (0.145) (0.095) (0.212) (0.175)** (0.159) (0.977) Foreign 0.066 0.377 -0.227 0.274 -0.057 0.252 (0.748) (0.253) (0.450) (0.229) (0.533) (0.202) Inverse Mill's ratio 0.082 0.482 0.097 0.448 0.087 0.527 (0.347) (0.310) (0.348) (0.313) (0.337) (0.311)* Country fixed effects Yes Industry fixed effects Yes Observations 1,293 1,410 1,293 1,410 1,293 1,410 Pseudo R-squared 0.15 0.15 0.15 0.16 0.15 0.15 Panel B. Pooled 2008 and 2005 samples Finance = Finance = Finance = Equity/assets Tier 1 capital Gains on fin assets Finance*Foreign -1.264 -0.319 -0.195 -0.004 -0.174 -0.078 (0.400)*** (0.196)* (0.122)* (0.178) (0.352) (0.215) Finance 0.024 0.006 -0.054 -0.386 -0.046 0.020 (0.095) (0.094) (0.161) (0.167)** (0.031) (0.035) Foreign 0.514 0.503 0.455 0.258 0.481 0.373 (0.197)*** (0.220)** (0.213)** (0.240) (0.267) (0.190)** Inverse Mill's ratio -0.777 -0.774 -0.774 -0.781 -0.644 -0.653 (0.098)*** (0.098)*** (0.098)*** (0.098)*** (0.084)*** (0.085)*** Country fixed effects Yes Industry fixed effects Yes Year fixed effects Yes Observations 1,664 1,664 1,664 1,664 1,710 1,710 Pseudo R-squared 0.14 0.13 0.13 0.13 0.13 0.13

Note: The dependent variable is a dummy variable equal to 1 if the firm is credit constrained. ‘Finance’ is a dummy equal to 1 if the respective financial variable from Table 4 is in the top 50% of its distribution. Each finance variable is locality-specific and is constructed by weighting equally (Columns 1, 3, and 5) or by number of branches (Columns 2, 4, and 6) the respective financial variable for each parent bank which has at least one branch or subsidiary in that locality. ‘Foreign’ is a dummy equal to 1 if the share of branches in each locality owned by branches or subsidiaries of foreign banks is more than 2/3, and to 0 if it is less than 1/3. ‘Inverse Mill’s ratio’ is the inverse of Mills’ratio from the probit model in Table 5 for each respective financial variable. The regressions also include the rest of the independent variables from table 6. The analysis is performed on all firms present in the 2008 survey (Columns (2), (4), and (6)) and on the subsample of firms present in both the 2008 and the 2005 survey (Columns (3), (5), and (7)). All regressions include country and industry fixed effects (Panel A), and country, industry, and year fixed efefcts (Panel B). *** indicates significance at the 1% level, ** at the 5% level, and * at the 10% level. See Appendix 1 for exact definitions. Source: BEEPS (2005 and 2008) and Bankscope (2005-2008).

Table 11. Probability of being constrained: differential effects

Panel A. Equally weighted bank data for each locality Tier 1 capital * External fin. dependence -0.071 (0.097) Tier 1 capital * R&D intensity -0.172 (0.107)* Tier 1 capital * capital intensity -0.134 (0.084)* City fixed effects Yes Industry fixed effects Yes Observations 1,226 1,226 1,226 Pseudo R-squared 0.15 0.16 0.16 Panel B. Branch-weighted bank data for each locality Tier 1 capital * External fin. dependence -0.144 (0.080)* Tier 1 capital * R&D intensity -0.185 (0.095)** Tier 1 capital * capital intensity -0.182 (0.080)** City fixed effects Yes Industry fixed effects Yes Observations 1,226 1,226 1,226 Pseudo R-squared 0.16 0.16 0.16

Note: The dependent variable is a dummy variable equal to 1 if the firm is credit constrained. ‘Finance’ is one of the three financial variables from Table 4. Each finance variable is locality-specific and is constructed by weighting equally (Panel A) or by number of branches (Panel B) the respective financial variable for each parent bank which has at least one branch or subsidiary in that locality. ‘External financial dependence’ is a dummy equal to 1 if the industry is in the bottom 50% of the distribution of industry medians of the proportion of capital expenditures financed with external funds for mature Compustat firms over the period 1990-2000. ‘R&D intensity’ is a dummy equal to 1 if the industry is in the top 50% of the distribution of industry medians of the ratio of research and development expenses to sales for mature Compustat firms over the period 1990-2000. ‘Capital intensity’ is a dummy equal to 1 if the industry is in the bottom 50% of the distribution of industry medians of capital usage per worker with external funds for mature Compustat firms over the period 1990-2000. ‘Inverse Mill’s ratio’ is the inverse of Mills’ratio from the probit model in Table 5 for each respective financial variable. The regressions also include the rest of the independent variables from table 6. The analysis is performed on all firms present in the 2008 survey. All regressions include city and industry fixed effects. *** indicates significance at the 1% level, ** at the 5% level, and * at the 10% level. See Appendix 1 for exact definitions. Source: BEEPS (2008) and Bankscope (2007 and 2008).

Appendix 1. Variables – Definitions and sources

Variable Name Definition Source Firm characteristics Small firm Dummy=1 if firm has less than 99 employees BEEPS 2005 & 2008 Medium firm Dummy=1 if the firm has betwee 100 and 499 employees BEEPS 2005 & 2008 Big firm Dummy=1 if firm has more than 500 employees BEEPS 2005 & 2008 Size Dummy=1 if the firm has more than 10 employees BEEPS 2005 & 2008 Age Dummy=1 if more than 10 years have elapsed since the firm’s incorporation BEEPS 2005 & 2008 Public company Dummy=1 if firm is a shareholder company / shares traded in the stock market BEEPS 2005 & 2008 Private company Dummy=1 if firm is a shareholder company / shares traded privately if at all BEEPS 2005 & 2008 Sole proprietorship Dummy=1 if firm is a sole proprietorship BEEPS 2005 & 2008 Privatized Dummy=1 if the firm went from state to private ownership in the past BEEPS 2005 & 2008 Subsidized Dummy=1 if the firm has received state subsidized in the past year BEEPS 2005 & 2008 Exporter Dummy=1 if firm’s production is at least partially exported BEEPS 2005 & 2008 Competition Dummy=1 if pressure from competitiors is “fairly” or “very” severe BEEPS 2005 & 2008 Audited Dummy=1 if the firm has its financial accounts externally audited BEEPS 2005 & 2008 Firm age Number of years since the firm’s year of incorporation BEEPS 2005 & 2008 Stand-alone firm Dummy=1 if the firm is stand-alone, 0 if it is part of larger establishment BEEPS 2005 & 2008 Manager experience Number of years of experience in the sector of the general manager BEEPS 2005 & 2008

41 Firm financing, credit demand, and credit access Investment – Earnings Share of recent investment financed by retained earnings, in % BEEPS 2005 & 2008 Investment - Bank credit Share of recent investment financed by bank credit, in % BEEPS 2005 & 2008 Investment - Trade credit Share of recent investment financed by trade credit, in % BEEPS 2005 & 2008 Needs loan Dummy=1 if the firm needs a loan BEEPS 2005 & 2008 Constrained Dummy=1 if the firm was refused a loan or didn’t apply for one because of adverse loan conditions BEEPS 2005 & 2008 Has loan Dummy=1 if the firm at present has a bank loan BEEPS 2005 & 2008

Industry benchmarks External financial dependence Median proportion of capital expenditures financed with external funds for mature Compustat firms Compustat over the period 1990-2000 R&D intensity Median proportion of the ratio of research and development expenses to sales for mature Compustat Compustat firms over the period 1990-2000 Capital intensity Median proportion of capital usage per worker for mature Compustat firms over the period 1990- Compustat 2000 Bank-level variables % foreign owned bank assets Share of banking sector assets owned by branches or subsidiaries of foreign banks EBRD Transition report 2008 Foreign Dummy=1 if more than 2/3 of the banking assets in the locality are owned by foreign banks, and to 0 Bankscope 2005-2008 if less than 1/3 are. Equity/assets Ratio of total equity to total assets Bankscope 2005-2008 Tier 1 capital Ratio of Tier 1 capital to total risk-weighted assets Bankscope 2005-2008 Gain on financial assets Gain on financial assets held by the bank Bankscope 2005-2008

42 Appendix 2. Domestic and parent banks in the sample

Country Bank Parent bank and country of incorporation Albania Alpha Bank Alpha Bank - Greece Raiffeisen Raiffeisen - Austira Banka Kombetare Trektare domestic Tirana Bank Pireus Bank - Greece Intessa San Paolo Bank Albania Intessa San Paolo - Italy National Bank of Greece National Bank of Greece Emporiki Emporiki Bank - Greece Banka Credins domestic Bulgaria Alpha bank Alpha Bank - Greece Unicredit Bulbank UniCredit Group - Italy DSK OTP - Hungary First Investment Bank domestic PostBank EFG Eurobank - Greece Expressbank Societe Generale - France United Bulgarian Bank National Bank of Greece Reiffeisen Raiffeisen - Austira Piraeus Piraeus Bank - Greece Croatia Zagrebaska Bank UniCredit Group - Italy Privredna Bank Zagreb Intessa San Paolo - Italy Erste & Steiermarkische Bank Erste Group - Austria Raiffeisen Bank Raiffeisen - Austria Societe Generale - Splitska Banka Societe Generale - France Hypo Alde Adria Bank Hypo Group - Austria OTP Banka Hrvatska OTP - Hungary Slavonska Banka domestic Hrvatska Postanska Banka domestic Czech Republic Ceska Sporitelna Erste Group - Austria CSOB KBC - Belgium Komercni Banka Societe Generale - France UniCredit Bank CR UniCredit Group - Italy Citibank Citibank - US Ceskomoravska zarucni a rozvojova banka domestic GE Money Bank GE Money - US Hypotecni Banka KBC - Belgium Raiffeisenbank Raiffeisen - Austira Estonia Swedbank Estonia Swedbank - Sweden SEB Skandinavska Enskilda Banken - Sweden Sampo Pank Danske Pank - Denmark Nordea Nordea Bank - Finland Hungary OTP Bank domestic K&H Commercial and Credit Bank KBC - Belgium MKB Bank Bayerische Landesbank - Germany CIB Bank Intessa San Paolo - Italy Raiffeisen Bank Raiffeisen - Austira Erste Bank Hungary Erste Group - Austria KDB Bank KDB Seoul - Korea UniCredit Bank Hungary UniCredit Group - Italy Latvia Parex domestic Hansabank Swedbank - Sweden Latvijas Krajbanka Snoras Bank - Lithuania SMP Bank domestic

43 Rietumu Banka domestic Trasta Komercbanka domestic Lithuania SEB Skandinavska Enskilda Banken - Sweden Sampo Pank Danske Pank - Denmark Nordea Nordea Bank - Finland Snoras Bank domestic Ukio Bankas domestic Hansabankas Swedbank - Sweden Parex Bankas Parex Group - Latvia Macedonia Alpha Bank Alpha Bank - Greece Stopanska Banka National Bank of Greece Komercijalna Banka domestic NLB Tutunska Banka NLB - Slovenia Ohridska Banka Societe Generale - France Pro Credit Bank Pro Credit Group Montenegro AtlasMont Bank domestic Crnogorska Komercijalna Banka OTP - Hungary Hypo-Alpe-Adria Bank Hypo Group - Austria Komercijalna Banka ad Budva domestic NLB Montenegro Banka NLB - Slovenia Prva Banka Crne Gore domestic Invest Banka Montenegro domestic Podgoricka Banka SG Societe Generale - France Opportunity Bank domestic Poland PKO Bank domestic Bank Pekao UniCredit Group - Italy Bank BPH UniCredit Group - Italy Bank Zachodni WBK domestic ING Bank Slaski ING Bank - Netherlands Bank Pocztowy domestic Kredyt Bank KBC - Belgium mBank Commerzbank - Germany Getin Bank domestic Romania BCR Erste Group - Austria BRD Group Societe General Societe Generale - France Volksbank Romania Volksbank - Austria Raiffeisen Bank Raiffeisen - Austira Alpha Bank Romania Alpha Bank - Greece UniCredit Tiriac Bank UniCredit Group - Italy Banca Transilvania domestic Bancpost EFG Eurobank - Greece CEC Bank domestic Slovakia Vseobecna Uverova banka Intessa San Paolo - Italy Slovenska Sporitelna Erste Group - Austria Tatra Banka Raiffeisen - Austira OTP Banka Slovensko OTP - Hungary Dexia Banka Slovensko Dexia - Belgium UniCredit Bank Slovakia UniCredit Group - Italy Volksbank Slovensko Volksbank - Austria CSOB Slovakia KBC - Belgium Slovenia Nova Ljubljanska Banka KBC - Belgium Nova Kreditna Banka Maribor domestic Abanka domestic SKB Societe Generale - France

44 UniCredit UniCredit Group - Italy Banka Koper Intessa San Paolo - Italy Banka Celje domestic Reiffeisen Krekova banka Raiffeisen - Austira